A fine-tuned large language model based molecular dynamics agent for code generation to obtain material thermodynamic...

Area of Science:

• Materials Science
• Computational Materials Science
• Artificial Intelligence in Science

Background:

• Materials science research increasingly uses AI-generated content (AIGC) for literature mining and data analysis.
• Automating theoretical calculations and code development in materials science remains a significant challenge.
• Existing methods require substantial manual effort for simulation program implementation.

Purpose of the Study:

• To propose a novel text-to-code generation approach for automating materials science simulation programs.
• To introduce the Molecular Dynamics Agent (MDAgent) framework for AI-driven code generation, execution, and refinement.
• To validate the effectiveness of automated code generation and review in thermodynamics simulations.

Main Methods:

• Utilizing large language models for text-to-code generation in materials science.
• Developing the Molecular Dynamics Agent (MDAgent) framework to guide AI models.
• Constructing a thermodynamic simulation code dataset for LAMMPS to fine-tune language models.
• Validating the approach with thermodynamics simulations using LAMMPS software.

Main Results:

• MDAgent framework successfully automates the generation, execution, and refinement of simulation code.
• Expert evaluation shows significant improvements in code generation and review capabilities.
• The proposed approach reduces average task time by 42.22% compared to traditional methods.
• A dedicated dataset for LAMMPS thermodynamic simulations was created for model fine-tuning.

Conclusions:

• AI-driven text-to-code generation offers a powerful solution for automating simulation program implementation in materials science.
• MDAgent framework enhances efficiency and accuracy in computational materials science tasks.
• This approach has significant potential to accelerate materials discovery and research.